FloodNet: A High Resolution Aerial Imagery Dataset for Post Flood Scene Understanding

Abstract

Visual scene understanding is the core task in making any crucial decision in any computer vision system. Although popular computer vision datasets like Cityscapes, MS-COCO, PASCAL provide good benchmarks for several tasks (e.g. image classification, segmentation, object detection), these datasets are hardly suitable for post disaster damage assessments. On the other hand, existing natural disaster datasets include mainly satellite imagery which have low spatial resolution and a high revisit period. Therefore, they do not have a scope to provide quick and efficient damage assessment tasks. Unmanned Aerial Vehicle (UAV) can effortlessly access difficult places during any disaster and collect high resolution imagery that is required for aforementioned tasks of computer vision. To address these issues we present a high resolution UAV imagery, FloodNet, captured after the hurricane Harvey. This dataset demonstrates the post flooded damages of the affected areas. The images are labeled pixel-wise for semantic segmentation task and questions are produced for the task of visual question answering. FloodNet poses several challenges including detection of flooded roads and buildings and distinguishing between natural water and flooded water. With the advancement of deep learning algorithms, we can analyze the impact of any disaster which can make a precise understanding of the affected areas. In this paper, we compare and contrast the performances of baseline methods for image classification, semantic segmentation, and visual question answering on our dataset.

Incidents1M: a large-scale dataset of images with natural disasters, damage, and incidents

“…Oftentimes, these studies are constrained to particular disaster events. However, recently deep learningbased techniques have been applied on larger collections of airborne or remote-sensed data to detect damaged buildings [43]- [48], segment flooded regions [49]- [51], estimate extent of fires [52], [53], assess hurricane destruction [54], [55], perform fine-grained analysis of disaster scenes [56], [57], and compute a disaster impact index [58]. Other studies have applied transfer learning [59] and few-shot learning [60] to deal with unseen situations during disasters.…”

Damage Estimation and Localization from Sparse Aerial Imagery

“…Increasingly, satellite and aerial imagery are being incorporated into post-disaster needs assessment [3]. However, while techniques exist for extracting information from orthorectified satellite and aerial imagery [4][5][6][7][8], methods for more general aerial imagery (such as oblique imagery from handheld cameras) have received far less attention. This is a critical limitation because post-disaster aerial imagery taken from handheld DSLR cameras from small, manned, fixed-wing aircraft remains popular due to the relatively low cost, high availability, conformity with existing regulations, and existing training programs associated with the practice [9,10].…”

“…Indeed, most available literature on registering oblique imagery relies on some variant of structure from motion or multiview stereo [19], which is the approach we take for georeferencing. For detecting damage, various deep learning approaches have achieved high accuracy in object detection challenges such as xView2 [4][5][6]. Previous work has also attempted to overcome lack of training data in either satellite or aerial images via transfer learning from satellite to aerial or vice versa [20][21][22].…”

Comprehensive Semantic Segmentation on High Resolution UAV Imagery for Natural Disaster Damage Assessment

“…The rescue team can use UAV (Unmanned Aerial Vehicle) to capture images of damaged properties and the whole affected area. Compared to satellite imagery, UAV imagery [3,4,5] provides higher resolution which helps to understand the detailed damage level of the captured area.…”